{
  "id": 74823,
  "title": "GPU runtime limits 2h",
  "url": "/competitions/quora-insincere-questions-classification/discussion/74823",
  "author_name": "",
  "post_date": "2018-12-16T07:14:06.874850300Z",
  "votes": null,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Sorry, i'm new to kaggle, and i want to know why there is a limit on GPU runtime when commit my kernel? This limit is 120min. \nThe reason i ask this is my kernel always exceed 120min when cross validation. I think CV is so important and i can't drop it.</p>",
  "messages": [
    {
      "id": "439719",
      "postDate": "12/16/2018 07:14:06",
      "content": "<p>Sorry, i'm new to kaggle, and i want to know why there is a limit on GPU runtime when commit my kernel? This limit is 120min. \nThe reason i ask this is my kernel always exceed 120min when cross validation. I think CV is so important and i can't drop it.</p>",
      "rawMarkdown": "Sorry, i'm new to kaggle, and i want to know why there is a limit on GPU runtime when commit my kernel? This limit is 120min. \nThe reason i ask this is my kernel always exceed 120min when cross validation. I think CV is so important and i can't drop it.",
      "votes": null
    },
    {
      "id": "439754",
      "postDate": "12/16/2018 09:15:58",
      "content": "<p>do your cv in a kernel that does not do model predictions. when satisfied with the CV results, make a kernel with only model predictions. problem solved :P</p>",
      "rawMarkdown": "do your cv in a kernel that does not do model predictions. when satisfied with the CV results, make a kernel with only model predictions. problem solved :P",
      "votes": null
    },
    {
      "id": "439783",
      "postDate": "12/16/2018 10:59:32",
      "content": "<p>The  sponsor chose this format  likely because it's a good regularizer for models complexity and endless ensembling. </p>\n\n<p>So,  optimisation and efficiency are part of the game :) </p>",
      "rawMarkdown": "The  sponsor chose this format  likely because it's a good regularizer for models complexity and endless ensembling. \n\nSo,  optimisation and efficiency are part of the game :)",
      "votes": null
    },
    {
      "id": "440056",
      "postDate": "12/17/2018 01:01:34",
      "content": "<p>Thank you so much! </p>",
      "rawMarkdown": "Thank you so much!",
      "votes": null
    },
    {
      "id": "440057",
      "postDate": "12/17/2018 01:07:09",
      "content": "<p>sorry, i didn't catch you, you mean one kernel do CV, and another kernel do model predictions? </p>",
      "rawMarkdown": "sorry, i didn't catch you, you mean one kernel do CV, and another kernel do model predictions?",
      "votes": null
    },
    {
      "id": "440209",
      "postDate": "12/17/2018 08:11:28",
      "content": "<p>If you are doing CV to find any hyper parameters, you can do that a separate notebook and after you find them desired parameters, just hardcode them in your submission notebook to cut down on computation time.</p>",
      "rawMarkdown": "If you are doing CV to find any hyper parameters, you can do that a separate notebook and after you find them desired parameters, just hardcode them in your submission notebook to cut down on computation time.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 439754,
      "author_name": "abhishek",
      "author_url": "",
      "post_date": "12/16/2018 09:15:58",
      "content": "<p>do your cv in a kernel that does not do model predictions. when satisfied with the CV results, make a kernel with only model predictions. problem solved :P</p>",
      "votes": null,
      "replies": [
        {
          "id": 440057,
          "author_name": "wuyaqiang",
          "author_url": "",
          "post_date": "12/17/2018 01:07:09",
          "content": "<p>sorry, i didn't catch you, you mean one kernel do CV, and another kernel do model predictions? </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 440209,
          "author_name": "weenkus",
          "author_url": "",
          "post_date": "12/17/2018 08:11:28",
          "content": "<p>If you are doing CV to find any hyper parameters, you can do that a separate notebook and after you find them desired parameters, just hardcode them in your submission notebook to cut down on computation time.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 439783,
      "author_name": "serigne",
      "author_url": "",
      "post_date": "12/16/2018 10:59:32",
      "content": "<p>The  sponsor chose this format  likely because it's a good regularizer for models complexity and endless ensembling. </p>\n\n<p>So,  optimisation and efficiency are part of the game :) </p>",
      "votes": null,
      "replies": [
        {
          "id": 440056,
          "author_name": "wuyaqiang",
          "author_url": "",
          "post_date": "12/17/2018 01:01:34",
          "content": "<p>Thank you so much! </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "439719": "Sorry, i'm new to kaggle, and i want to know why there is a limit on GPU runtime when commit my kernel? This limit is 120min. \nThe reason i ask this is my kernel always exceed 120min when cross validation. I think CV is so important and i can't drop it.",
    "439754": "do your cv in a kernel that does not do model predictions. when satisfied with the CV results, make a kernel with only model predictions. problem solved :P",
    "439783": "The  sponsor chose this format  likely because it's a good regularizer for models complexity and endless ensembling. \n\nSo,  optimisation and efficiency are part of the game :)",
    "440056": "Thank you so much!",
    "440057": "sorry, i didn't catch you, you mean one kernel do CV, and another kernel do model predictions?",
    "440209": "If you are doing CV to find any hyper parameters, you can do that a separate notebook and after you find them desired parameters, just hardcode them in your submission notebook to cut down on computation time."
  },
  "source": "meta"
}